James Gates sues OpenAI, alleging ChatGPT helped his wife plan his poisoning

ChatGPT should have interrupted an apparent poisoning plot, not helped develop it, a newly public lawsuit argues. James Gates, a North Carolina man, is suing OpenAI and CEO Sam Altman, alleging the chatbot reinforced his wife’s suspicions about him and supplied guidance for an attempt on his life. Those allegations have not been established in court.

The civil complaint was filed in San Francisco County Superior Court. Gates’s lawyers at Tech Justice Law say they and co-counsel filed it on October 14, 2026; Ars Technica reported that it became public this week. The case targets both the chatbot’s responses and the safeguards Gates alleges should have stopped them.

From self-protection questions to alleged attack planning

Gates initially recommended ChatGPT to his wife, Cheryl, to help prepare reusable templates for her work as a pediatric occupational therapist. According to the complaint, she soon used it for everyday family tasks as well, including children’s activities, medication questions and marital advice.

The suit alleges that the chatbot amplified her fears about a possible divorce and adopted an unfounded portrayal of her husband as dangerous and controlling. CNET identifies the model involved as GPT-4o and reports that the complaint describes it praising Cheryl while echoing her negative descriptions of James.

Her initial poisoning questions were framed as self-protection: she said she feared her husband might harm her. The complaint alleges that ChatGPT continued answering as the questions became more explicitly threatening. It describes at least 25 exchanges about poisoning, including personalized guidance and advice on concealing substances in drinks.

Gates also alleges that ChatGPT falsely assured Cheryl that deleting conversations would make them impossible to recover, even through a subpoena or court order. His lawyers characterize the exchanges as personalized assistance in planning and concealing violence, rather than the passive delivery of publicly available information.

An illness, a handwritten note and disputed criminal-case details

Gates alleges that Cheryl repeatedly tried to poison him between June and September 2025. During the most serious episode, he says, he was ill for 72 hours, with cardiac symptoms, abdominal pain, weakness and profound confusion. He later found a handwritten note listing substances and instructions to clear conversations and log out of ChatGPT.

According to the complaint, the note prompted him to contact police, whose investigation uncovered the chat logs. Gates says the alleged poisoning has left him with ongoing emotional distress, PTSD, insomnia and economic harm. These are claims in his civil case, not judicial findings about what caused his illness.

The criminal-case accounts differ materially. Ars Technica reports an October 2026 grand-jury indictment for attempted first-degree murder and food contamination. CNET, citing USA Today, says Cheryl was charged in October 2025 and those charges were later dropped for insufficient evidence. The two accounts do not establish a consistent current criminal-case status.

Safety is a top priority. We banned this user and have provided information to law enforcement as part of their investigation into this case.

An OpenAI spokesperson, in a statement to Ars Technica

Gates wants intervention before a conversation becomes a plan

The civil case includes product-liability claims alleging defective design and failure to warn, negligence and violations of California’s unfair competition law, according to Gates’s lawyers. It asks the court for more than compensation: Gates wants changes to how ChatGPT handles conversations that appear to involve planning violence against identifiable people.

His complaint argues that commercially feasible moderation could have detected the alleged pattern. Rather than continue responding, he says, the chatbot should have ended the exchanges or directed Cheryl toward law enforcement, therapeutic services or poison-control help. The alleged failure is therefore not just one harmful answer, but continued assistance across conversations.